Aspiration Levels and Educational Choices
An experimental study
Lionel Page
∗
Louis Levy Garboua
†
Claude Montmarquette ‡
October 2006
∗
Westminster Business School, University of Westminster, 35 Marylebone Road, NW1 5LS, London,
[email protected]
†
Center of Economics Sorbonne, 106-112 Bd de l’Hôpital, 75647 Paris Cédex 13 and CIRANO, Montreal, Canada. [email protected]
‡
CIRANO and Université de Montréal, 2020 rue University, Montreal, (Quebec), Canada, H3A 2A5.
[email protected].
1
Abstract
The explanation of social inequalities in education is still a debated issue in
economics. Recent empirical studies tend to downplay the potential role of credit
constraint. This article tests a different potential explanation of social inequalities
in education, specifically that social differences in aspiration level result in different
educational choices. Having existed for a long time in the sociology of education,
this explanation can be justified if aspiration levels are seen as reference points in a
Prospect Theory framework. In order to test this explanation, this article applies the
method of experimental economics to the issue of education choice and behaviour.
One hundred twenty-nine individuals participated in an experiment in which they
had to perform a task over fifteen stages grouped in three blocks or levels. In order
to continue through the experiment, a minimum level of success was required at the
end of each level. Rewards were dependent on the final level successfully reached. At
the end of each level, participants could either choose to stop and take their reward
or to pay a cost to continue further in order to possibly receive higher rewards.
To test the impact of aspiration levels, outcomes were either presented as gains
or losses relative to an initial sum. In accordance with the theoretical predictions,
participants in the loss framing group choose to go further in the experiment. There
was also a significant and interesting gender effect in the loss framing treatment,
such that males performed better and reached higher levels.
Keywords : Education inequality, Prospect Theory, Experimental
Economics
JEL Classification : I21, D80, J24, C91
2
The effects of social background on educational achievement are well documented. Empirical data show that children from higher socio-economic backgrounds perform better at
school and choose longer courses of study than their counterparts from lower backgrounds
(Duru-Bellat 2002). Numerous studies have increased our understanding of these phenomena. The precise mechanisms underlying these differences are, however, still a research
issue. In economics, the possible causes of those facts are considered to be respectively the
transmission of human capital and the existence of credit market imperfection. However,
how much these factors explain social inequalities in education is still an issue. Recent
studies by Carneiro and Heckman (2002) and Cameron and Taber (2004), for example,
tend to downplay the role of credit constraints in educational choice. Another kind of
explanation has been proposed in sociology of education. Since Boudon (1973), sociologists have tried to explain social inequalities in education as a result of initial differences
in aspiration levels. This explanation has not appeared in the economic literature, the
notion of aspiration being, a priori, somewhat alien to the notion of preference.
Page (2005a, 2005b) has, however, shown that the impact of aspiration levels on educational outcomes can be modeled with the notion of reference point from prospect theory.
In his theoretical framework, the preferences of the participant are characterised by a reference point, which divides the space of outcomes between outcomes perceived as losses,
and outcomes perceived as gains. This reference point affects the behaviour of participants. Individuals will have a tendency to be risk seeking when presented with lotteries
with outcomes defined as losses. Participants will also tend to subjectively overweight
losses as compared to gains (loss aversion). Given these effects of the reference point on
participant utility, it is possible to show that participants with a higher reference point
will have a tendency to choose longer studies given a more risk seeking behaviour. Furthermore, they will expend more effort in order to increase their probability of success in
education due to a higher marginal utility resulting from loss aversion.
It is difficult to separate empirically the predictions of prospect theory from other
explanations because reference points or aspiration levels are mainly unobservable. Moreover, when they can be approximated, the reference points are likely to be endogenous
3
since they may depend on all the unobserved variables that affect educational success. The
usual problem of non experimental data - the multiplicity and intricacy of causal variables
and the persistent doubt of endogenous bias - makes decisive conclusions difficult in this
context. Thus, the experimental methodology may be a worthwhile complement to the
usual field studies. The simplified framework of an economic model may be reproduced
within an experimental design. It is then possible to control each variable and to exogenously modify the variables of interest. In particular, the literature on prospect theory
has shown that it is possible to manipulate the reference point of participants by framing
the outcomes of the experimental study differently.
In our experiment, we replicate the relevant features of educational choices found in
human capital investment models. Feature of these models include investing both time
and money to perform laborious training tasks in order to get monetary rewards. Tasks
in this experiment are grouped in stages, which require some minimal levels of success to
proceed to the next stage. Moreover twice during the experiment (at stage 9 and 12),
participants can chose to stop and have their current gains or continue and have a chance
to win more. The choice to continue is however costly and the result of this investment is
uncertain. It depends of the success of the individual in the later stages of the experiment.
It is obviously impossible in an experimental setting to reproduce all the characteristics of an educational choice that imply high stakes and long term consequences. The
explanation of social inequalities using differences in aspiration levels does not however
rely on the size of the stakes or on the time dimension of the choice. Our experiment aims
therefore at reproducing the kind of simple modeled situations where aspiration levels
have been proposed to have an effect in order to assess the predicted impact of differences
in aspiration levels.
The experiment is made of two treatments. In one treatment, the outcomes are displayed as gains, framing a low reference point. In the other treatment,the outcomes are
presented as losses, framing a high reference point. According to prospect theory, the
framing of the monetary outcomes as losses should have two effects:
(i) The participants should be more likely to choose to continue at stages 9 and 12.
4
(ii) The participants should exert more effort to perform the task.
In accordance with the predictions of prospect theory, we find in our experiment that
participants with a high reference point choose more often to continue longer in the
education game than those with a low reference point. The high reference point framing
also seems to have an effect on the level of effort. However, surprisingly this effect seems
to appear only for males who improve their performance in the face of losses. In contrast,
the performance of females does not differ between the two treatment conditions or even
gets lower in the loss framing treatment.
1
Experimental Design
Experiment structure
The experiment consists of 15 stages grouped in 3 levels. Each stage involves solving a
given number of anagrams. The first level contains the stages 1 to 9, the second level
the stages 10 to 12 and the third level the stages 13 to 15. At the end of each level, a
participant must have solved two thirds of the anagrams to be allowed to pass to the next
level. The difficulty of the level increases according to the following criteria:
• The number of anagrams per stage increases with the level with a constant time
limit of 8 minutes per stage. Specifically: .
– 6 anagrams per stage for level 1,
– 9 anagrams per stage for level 2 and
– 12 anagrams per stage for level 3
• The length of anagrams increases on average.
The structure of the experiment is represented in Figure 1. At the end of each level, the
participant fails or passes, and correspondingly there are two possible outcomes in terms
of monetary payments.
5
First Level
1
2
8
Second Level
9
10
11
Third Level
12
13
14
15
Figure 1: Structure of the experiment
Framing of the monetary payments
The monetary payments of this experiment depend on the highest level successfully
reached. The experiment is composed of two treatments corresponding to two distinct
levels of aspiration. The aspiration level is affected by the framing of the payments. In
the first treatment, the payments are presented as gains. In the second treatment, the
participant is told that he or she receives an initial sum. Stopping at early stages implies
losing part of this sum. Framing the payment outcomes as gains gives implicitly to the
subjects a natural reference point equal to zero. Each increase in the monetary outcome
is then perceived as an increase in gains. On the contrary, framing the payment outcomes
as losses relative to an initial sum gives a reference point at the level of this sum. Each
increase in the monetary outcome is perceived as a decrease in losses. The structure of
payments is actually the same, but the presentation differs (see Table 1)1 .
Table 1: Payments framing for the two treatments
Level successfully
passed
Initial endowment
none
1
2
3
Treatment “Gains”
none
$2
$10
$20
$35
Treatment “Losses”
$35
-$33
$-25
$-15
$0
1
The outcomes in this paper are in Canadian dollars and correspond to the first sessions of the
experiment conducted in Montreal.
6
Participant choices
At the end of levels 1 and 2, the participants who solved at least two thirds of the
anagrams can choose to continue or to stop. To continue has a cost, which represents the
opportunity and the education costs of studies. With successful completion of the level,
the participant gets a higher reward, but with failure she will receive a lower monetary
outcome than if she had chosen to stop. After stage 9, to continue to level 2 costs $6
and after stage 12, to continue to level 3 costs $9. It is possible to represent the decision
tree of the participant to show the risky choices. Figure 2 shows the two risky decisions
facing the participant during the experiment. At stage 9 and 12, the participant has to
chose between a risky lottery linked to the choice of continuing (Success, p; Failure, 1 − p)
and a sure outcome linked to the choice to stop. For each of these choices, we suppose
that the participant estimates a subjective probability of success respectively p1 and p2
before to continue or to stop. For example, at the end of the stage 9, if the participant
solved less than two thirds of the anagrams of stages 1-9, he or she is eliminated and get
2$. Otherwise, the participant can choose to stop and get the sure amount of 10$, or try
to continue for the cost of 6$. In case of success, with probability p1 he or she would
get 20$, but in case of failure, with probability 1 − p1 , the final gain would only be 4$.
The participants are thus faced with choices between a sure amount and a risky lottery
at stages 9 and 12.
1.1
Experimental sessions
There were a total of eight sessions (four per treatment) with 129 participants. Four
sessions were conducted in the Bell laboratory in Montréal in February 2006, and four
sessions were conducted in the University of Paris 1 laboratory of experimental economics
in April 2006. The participants were contacted either by email2 or by advertising on the
research center websites. Participants did not know a priori the topic of the experiment
but were informed of the expected duration and the existence of a reward. The two
2
Emails were sent to a database of individuals having previously registered their interest into participating to an experiment.
7
9
failure
2
Level sucessful
10
(20,p1 ;4, 1 − p1 )
12
failure
4
Level sucessful
20
(35,p2 ;11, 1 − p2 )
15
failure
Level sucessful
11
35
Figure 2: Decision tree
treatments (Gain Framing and Loss Framing) were performed in Paris and Montreal
(with a roughly equal number of participants for each treatment in each laboratory).
During the experiment, each participant was given a participation fee of $10 or e8 and
randomly assigned to a computer. They all received standard instructions on the nature
of the experiment and the task required. Before the experiment, the participants were
asked about their age, gender, native language3 , education level, job status, the frequency
they played Scrabble or did crosswords. They were tested regarding their comprehension
of the instructions. At the end of the experiment, the participants were asked about their
level of stress, motivation to solve anagrams, and level of satisfaction with their results.
2
Experimental Results
2.1
Descriptive Statistics
Comparing the initial characteristics of participants between treatments (Table 2) no
significant differences can be noted. The random selection of participants seems to have
3
All the anagrams and more generally the experiment were in French.
8
provided two samples fairly similar relative to characteristics which could play a role in
the ability to succeed in the experiment. In order to measure potential difference in
individual risk aversion, participants were asked if they would prefer to choose between
a sure amount of 5$ or to pick a ball in box with white and black balls whose respective
proportions are unknown and get 10$ if the ball picked is black. The binary answer to
this question is used here as a rough index of individuals’ risk aversion. No significant
difference in the answer to this question was noticed.
Male
Age
Previously participated to an experiment
Not a native French speaker
Frequency of practice of crosswords
Frequency of practice of scrabble
Risk averse answer to hypothetical lottery question
N
a
Gain Framing
b
GFa
LFb
Diff
57%
25.3
67%
26%
2
2
46%
53%
25.5
64%
32%
2.5
2.2
42%
ns
ns
ns
ns
ns
ns
ns
63
66
Loss Framing
Table 2: Sample characteristics
Figure 3 illustrates the “surviving” profile of our experiment. One major goal of the
design was to have a progressive selection of participants. Due to failure and self-selection,
the number of participants is bound to decrease at each level. The “surviving” profile of
the experiment is satisfactory. The difficulty of the task ensure that a real selection takes
place along the experiment. However, the task is not too hard, and a reasonable number
of participants manage to finish it successfully.
9
100
Percentage of successful individuals for each level
83.33
Percents
40
60
80
85.71
37.88
30.15
20
18.19
0
9.52
Level 1
Level 2
GF
Level 3
LF
Figure 3: Survival rates at each level
2.2
Differences in choices
The main goal of our experiment was to study how the framing of a high reference point
affects choices to continue from one level to another. Figure 4 shows the differences in
choices made by participants in both treatments. The participants from the LF group
tend to choose to continue more often than the participants from the GF group. These
differences are significant at the 10% level for a one tail test at stage 9 (level 1) and at 5% at
stage 12 (level 2). This result fully supports the theoretical assumption that participants
with higher reference points will be more risk seeking. A joint non parametrical test
(Kruskal Wallis rank test) on both levels indicates that when the results of those two levels
are taken jointly, the level of statistical significance of the difference in choice reaches 5%
for a two tail test.
Of particular interest, males tend to choose to continue more often than females. Both
males and females in the LF treatment tend to continue more than in the GF treatment.
The econometric analysis of the factors determining the choice to continue or not at
stages 9 and 12 confirmed the results of the descriptive statistics. In Table 3, a probit
model (with 1 for those continuing in later stages and 0 otherwise for the observed variable)
10
0.97
0.91
0.83
0.68
0
.2 .4 .6 .8
1
Stage 9
Females
Males
GF
LF
0.78
0.58
0.57
0.29
0
.2 .4 .6 .8
1
Stage 12
Females
Males
GF
LF
Figure 4: Individual choices to continue after stage 9 and 12
indicates that participants in the LF group show a higher probability to continue further at
stage 9. A one tail test on the coefficient of the LF dummy indicates that, in accordance
with the prediction, it is positive at 5% level in the regressions (2) and (3). The non
significance of the result in regression (1) may be due to the necessity to control for
other variables. The fact of introducing relevant variables in the regression increases the
coefficient and its significance. The introduction of additional explanatory variables for
the choice at stage 12 also increases the coefficient, but the latter is no longer significant
at 5%. In regressions (5) and (6) the observations of choices are pooled on both stage
9 and 12 and a dummy controlling for the given level is introduced. The coefficient of
the LF dummy is significant at 5% for a one tail test in regression (5) and for a two
tail test for the regression (6). In regressions (2), (4) and (6) of Table 3, we control for
participants’ characteristics such as gender, native language, and the previous practice of
scrabble. We also controlled for the ability of the participant at the task using the mean
time per anagram solved during the stages 1 to 9 for the choice at level 1 and the mean
time per anagram solved during the stages 10 to 12 for the choice at level 2. The binary
variable giving an indication about the risk aversion of the individuals is introduced in
11
Table 3: Choices : probit regressions
Choice stage 9 Choice stage 12 Choice both stages
LF
(1)
0.439
(2)
0.693
(3)
0.649
(4)
0.701
(5)
.436
(6)
.656
(1.42)
(1.77)†
(1.65)†
(1.53)
(.232)†
(.061)∗
Male
Not French Native
0.785
0.817
∗
†
(2.01)
(1.73)
(.069)∗∗
-0.963
-0.247
-.727
(0.35)
(.095)∗
∗
(2.37)
Play Scrabble
Abilitya
Risk aversionb
.819
-0.208
0.114
-.046
(1.45)
(0.75)†
(.019)
-0.027
-0.031
-.046
(2.99)∗∗
(1.88)†
(.001)∗∗
-1.092
-1.126
-1.079
∗
∗
(.067)∗∗
(2.56)
(2.31)
Dummy Level 12
-1.005
(.085)∗∗
Constant
0.896
∗∗
Observations
3.092
∗∗
-0.066
∗∗
(4.53)
(3.66)
(0.23)
109
108
44
1.334
(1.17)
43
.601
∗∗
(.156)
153
2.613
(.626)∗∗
151
Absolute value of z statistics in parentheses
Significant : † 10% ; ∗ 5%; ∗∗ 1%
a
Ability is measured with the mean time individual required to solve one anagram at
the previous level.
b
Dummy equal to 1 if the participant choose an uncertain lottery in a hypothetical choice
regressions (2), (4) and (6). The coefficients of this variable in these probit regressions
indicate that the participants who made a risk averse choice tend also to be more risk
averse during the experiment. Importantly, these individual differences in risk aversion
do not account for the differences in choices between treatments.
The coefficient of the LF treatment dummy variable is important. The corresponding
marginal effect is around 9 percentage points at level 1, 25 at level 2 and 13 for both levels
taken together. Thus, the participants in the LF treatment group tend to take greater
risks by continuing more often when they are faced with the choice to leave or to continue.
Therefore, the framing does matter.
12
2.3
Gender differences in performances
Our second proposition is that participants in the LF treatment should perform better at
the task than the participants of the GF treatment due to a higher motivation to expand
effort. At first sight, no sign of such difference can be found. Figure 5 shows that the
participants from both treatments can’t be distinguished relative to their performance on
average.
Cumulated number of anagrams
0
2
20
4
40
6
60
8
80
10
100
Number of anagrams solved
1
2
3
4
5
6
7
8 9
Stages
GF
10 11 12 13 14 15
1
2
3
4
5
6
LF
7
8 9
Stages
GF
LF
Mean time per anagram
20
100
40
200
60
300
80
400
100
Time spent per stage
10 11 12 13 14 15
1
2
3
4
5
6
7
8 9
Stages
GF
10 11 12 13 14 15
1
LF
2
3
4
5
6
7
8 9
Stages
GF
10 11 12 13 14 15
LF
Figure 5: Average performance at each stage for both treatments
This general picture masks very important gender differences within groups. The main
result of our experiment regarding gender is that the LF framing has a clear positive effect
on performance, but only for males. Figure 6 shows the results by gender for two of the
main indicators of performance of the experiment, Number of anagrams solved, and mean
time to solve an anagram, which were represented in Figure 5. While no gender differences
are noticeable in the GF treatment, males systematically outperform females in the LF
treatment. On average males solve more anagrams than females at each stage in the
LF treatment, and they take less time per anagram in 14 of the 15 stages. A Wilcoxon
signed rank test shows that these differences are significant at 1% for both indicators.
These results can’t be attributed to differences in the characteristics of males and females
13
in both treatment as no significant difference were found between the characteristics of
males in both treatments and between the characteristics of females in both treatments.
8
6
4
2
2
4
6
8
10
Number of anagrams solved in the LF
10
Number of anagrams solved in the GF
1
2
3
4
5
6
7
8 9
Stages
Females
10 11 12 13 14 15
1
2
3
4
5
Males
6
7
8 9
Stages
Females
Males
100
50
0
0
50
100
150
Mean time per anagram in the LF
150
Mean time per anagram in the GF
10 11 12 13 14 15
1
2
3
4
5
6
7
8 9
Stages
Females
10 11 12 13 14 15
1
Males
2
3
4
5
6
7
8 9
Stages
Females
10 11 12 13 14 15
Males
Figure 6: Gender differences in the GF and LF treatments
This gender difference in performance has consequences relative to the final outcome
of participants. Figure 7 shows how differently males and females perform when in the LF
treatment. While no difference is noticeable in the GF treatment, males reach on average
a higher level in the experiment and earn an additional 7$ relative to females (both results
are significant at 10%).
Our data allows to study a bit further the psychological states of males and females in
both treatments. In particular, an explicit question at the end of the experiment aimed
at eliciting the level of motivation of the participant. The question was “How would you
describe your motivation to solve the anagrams?”. The participants were invited to answer
on a Likert scale fron 1 (very low) to 7 (very high). Figure 8 shows that while no gender
difference is noticeable in motivation in the GF treatment, males in the LF treatment
report being more motivated than females to solve the anagrams at a 10% significance
level.
Solving anagrams requires sustained concentration, and this is where an effort must be
14
Gains in the experiment
3
.5
6
1
Level reached
Canadian Dollars
9
12
1.5
15
2
Level reached successfully
GF
LF*
GF
Treatment group
Females
LF*
Treatment group
Males
Females
Males
Gender differences significant at ** 5\%, *** 1\%
Figure 7: Individuals final outcome
made. Two indirect indicators of concentration were also assessed. First, at the beginning
of the experiment, after the reading of the experiment rules to participants, an initial
questionnaire was given to test the understanding of the experiment by every participant.
In case of a wrong answer, an explanation was given to ensure full understanding, and the
participant could change his or her answer. Each participant had to answer all questions
correctly to be allowed to proceed further in the experiment. Several factors can play on
the probability to make a mistake, including the concentration of the participant during
the preliminary phase of explanation. An interesting result is that while no difference is
noticeable in the GF treatment, females make more mistakes in this initial questionnaire
than males at a 10% level of significance4 . A second indirect indicator of the level of
concentration during the experiment is the level of stress of the participant during the
task. A question was asked about their level of stress during the experiment (between 1 to
7). While no gender differences appear in the GF treatment, males seem to have a higher
4
Differences in understanding between males in females in the LF treatment could then be thought
as a reason for differences in performance. This does not seem however to be a likely explanation.
Participants making mistakes had the good answer reexplained to them, reducing then the understanding
differences. Moreover differences in the understanding of the instructions can hardly explain differences
in performances to solve anagrams.
15
level of stress in the LF treatment (this difference is however not statistically significant).
Satisfaction
4.5
Motivation
5
5.5
Level of satisfaction
3
3.5
4
4.5
6
Motivation to solve anagrams
GF
LF*
GF
Treatment group
Males
Females
Stressfulness of the experiment
3
3.5
4
4.5
Mistakes in the initial questionnaire
1.2 1.4 1.6 1.8 2
Females
Mistakes
GF
LF*
Treatment group
Females
LF***
Treatment group
Males
Males
Stress
GF
LF
Treatment group
Females
Males
Gender differences significant at * 10\%, ** 5\%, *** 1\%
Figure 8: Gender differences in the two treatments
The answers to the final questions seem then to indicate that males where more motivated during the experiment and expended a higher level of effort in order to solve the
anagrams. An important feature is that the gender differences seem to be not only due
to an improvement of the males’ performance, but also, as shown on figures 7 and 8, to a
decrease in the performance of females in the LF treatment5 .
Eventually, when asked about their satisfaction about their performance in the experiment, a very significant gender difference appear in the LF treatment (1% level), males
declaring to be more satisfied than females, while no difference was noticeable in the GF
treatment (Figure 8). Importantly this difference in satisfaction is not due to differences
in gains. When controlling by regression for the fact that males in the LF treatment earn
more at the end of the experiment the difference in satisfaction between males and females
in the LF treatment group is still significant at 5%.
5
This phenomenon is however too slight to be statistically significant with our number of observations.
The repetition of this pattern for the different variables is however striking
16
3
Discussion
According to numerous sociologists of education, aspiration levels matter and are a major explanation of social inequalities in education. Our experiment aimed to study how
aspiration levels, defined as reference points, influence individual choices in a situation
designed to replicate educational choices. A very important result of this experiment is
that framing outcomes as losses relative to a high initial amount induces participants to
chose to continue further and to take more risks.
This situation may be easily understood in terms of real educational choices. For
example, imagine two individuals with different social backgrounds. Emma is from a poor
social background whose parents have a low level of education, while Ben is from a high
social background whose parents have a very high level of education. If social background
and the education level of parents act as a reference point, then Emma and Ben will have
different educational aspirations. For Emma, any additional level of education beyond
that of her parents will be perceived as a positive achievement. In contrast, for Ben, the
same level of education would be perceived as a failure if it is lower than his parents’ level
of education. While Emma would consider her outcome as positive if stopping at any
intermediate level of education, Ben would consider the same outcome as negative.
Our experiment aimed to reproduce the situations of Emma and Ben, respectively in
our GF and LF treatments. In accordance with the theoretical predictions, the individuals
from the LF treatments, with a high reference point, choose more often to continue in
the experiment. Therefore, this supports the idea that aspiration levels may play a major
role in educational choices, and that social differences in aspiration levels may be a major
cause of the social inequalities in educational outcomes.
The differential effect of the LF treatment for males and females was not initially
expected from our experimental design. It is, however, not fully surprising relative to
previous studies. A very relevant finding which can help explain this result is from Rizzo
and Zeckhauser (2003) who found that males and females with high reference points do
not react in the same way, with males being more likely to change their behaviour in order
to reach a higher goal than females. Studying the behaviour of male and female doctors,
17
they found that 1) generally males have higher reference points that females, and 2) given
males and females with the same high reference point, males will be more likely to modify
their behaviour to reach their goal. In this study the authors found that male doctors
with high reference points are more likely to take a higher work load in order to reach
their high income reference point.
In comparison to this study, we found first that males are more likely to choose to
continue in the experiment in both treatments. This could be explained by a tendency to
adopt a higher reference point than females in each treatment6 . It must be stressed that
no significant gender difference was recorded for the question on risk aversion, therefore
gender difference in choices do not seem to stem from gender differences in risk aversion
per se. Second, while males and females have similar performances in the GF treatment,
males outperform females in the LF treatment. This is consistent with the second result of
Rizzo and Zeckhauser (2003) where the framing effect primarily affects males’ behaviour.
Whilst the effect of the framing on performances of males and females seems relatively
minor on Figure 6, the higher performance of males in the LF condition leads to very
noticeable over representation of males at the highest level of success in the experiment.
As shown in Table 4, males from the LF treatment represent 55% of the participants
reaching the highest level successfully compared to 25% which would be expected from
chance alone.
Table 4: Distribution of participants successfully completing the experiment
Females Males
GF
LF
2
2
4
10
On the whole, males represent 78% of the highest achievers while they represent only
55% of the participants. This phenomenon led us to wonder about the possible comparison
with gender differences in reality. A well established phenomenon is the concentration of
males in high levels of education while females are actually performing on average better in
6
Our design aims at framing a reference point of 0 in the GF treatment, and of 35$ in the LF treatment.
However, it is more than likely that the framing effect does not fully determine the aspiration level of the
participants.
18
previous stages (Davies and Guppy 1997). Our results raise an interesting yet unanswered
question : “could the concentration of males in higher levels of education be due to the
higher rate of success of males with high aspiration levels ?”
4
Conclusion
This paper presents an experiment reproducing the main features of educational choices.
By framing the outcomes either as gains or as losses relative to the highest possible
outcome, it aims to study how aspiration levels conceived as reference points in a prospect
theory framework may influence educational choice. Sociologists have first proposed that
a major part of social inequalities in education be due to social differences in aspiration
level.
The results of the experiment are in accordance with the prospect theory predictions
concerning the effect of aspiration levels on choices. We find that to frame outcomes as
gains or losses in our experiment significantly changes the choices of the participants.
Participants in the loss framing treatment chose more often to continue further in the
stages of the experiment than participants in the gain framing treatment. Concerning
the effect of aspiration levels, the prediction stemming from prospect theory are only
validated for males. The framing of outcomes as losses, which was expected to increase
the motivation of the participants, does so, but only for males.
Two comments about these results. First, that a simple framing of the outcomes may
reproduce inequalities in choices and to some extent in motivation is of clear interest for
the explanation of social inequalities in education. This result supports the idea that
differences in aspiration levels play a role in social inequalities in education.
Second, the outcome difference in gender behaviour in the loss framing treatment deserves further work given that gender inequality in education outcomes is still an empirical
issue. In our experiment, not only do the males in the LF treatment report a higher level
of motivation, but they are also more likely to reach a higher level. It seems that gender
inequality in education could at least partly stem from the effect of aspiration levels as
shown in our experiment.
19
References
Boudon, R. (1973): L’inégalité des chances. Armand Colin, Paris.
Cameron, S., and C. Taber (2004): “Estimation of Educational Borrowing Constraints
Using Returns to Schooling,” Journal of Political Economy, 112(1), 132–182.
Carneiro, P., and J. Heckman (2002): “The evidence on credit constraints in postsecondary schooling,” Economic Journal, 112(482), 705–734.
Davies, S., and N. Guppy (1997): “Fields of Study, College Selectivity, and Student
Inequalities in Higher Education,” Social Forces, 75(4), 1417–1438.
Duru-Bellat, M. (2002): Les inégalités sociales à l’école. Puf, Paris.
Page, L. (2005a): “Motivation and Goal Setting, a Prospect Theory Model for Risky
Situations,” Mimeo.
(2005b): “Social inequalities in educational choices and Prospect Theory,”
Mimeo.
Rizzo, J., and R. Zeckhauser (2003): “Pushing Incomes to Reference Points – Why
male doctors earn more,” Operation Research, 30.
20